|Publication number||US8005104 B2|
|Application number||US 12/815,928|
|Publication date||Aug 23, 2011|
|Filing date||Jun 15, 2010|
|Priority date||Jul 8, 1998|
|Also published as||US6967962, US7477652, US7738474, US20050232288, US20090109989, US20100246595|
|Publication number||12815928, 815928, US 8005104 B2, US 8005104B2, US-B2-8005104, US8005104 B2, US8005104B2|
|Inventors||Eitan Medina, David Shemla|
|Original Assignee||Marvell Israel (M.I.S.L.) Ltd.|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (16), Non-Patent Citations (2), Classifications (24), Legal Events (1)|
|External Links: USPTO, USPTO Assignment, Espacenet|
This application is a continuation of U.S. patent application Ser. No. 12/348,457, filed Jan. 5, 2009, (now U.S. Pat. No. 7,738,474) which is a continuation of U.S. patent application Ser. No. 11/152,901, filed Jun. 15, 2005 (now U.S. Pat. No. 7,477,652), which application is a continuation of U.S. patent application Ser. No. 09/348,865, filed Jul. 7, 1999 (now U.S. Pat. No. 6,967,962). This application claims the benefit of Israel Application No. 125,273, filed Jul. 8, 1998. The disclosures of the above applications are incorporated herein by reference in their entirety.
The present invention relates to network switches generally and to cross-bars in particular.
A network switch creates a network among a plurality of end nodes, such as workstations, and other network switches connected thereto. Each end node is connected to one port of the network. The ports also serve to connect network switches together.
Each end node sends packets of data to the network switch which the switch then routes either to another of the end nodes connected thereto or to a network switch to which the destination end node is connected. In the latter case, the receiving network switch routes the packet to the destination end node.
Each network switch has to temporarily store the packets of data which it receives from the units (end node or network switch) connected to it while the switch determines how, when and through which port to retransmit the packets. Each packet can be transmitted to only one destination address (a “unicast” packet) or to more than one unit (a “multicast” or “broadcast” packet). For multicast and broadcast packets, the switch typically stores the packet only once and transmits multiple copies of the packet to some (multicast) or all (broadcast) of its ports. Once the packet has been transmitted to all of its destinations, it can be removed from the memory or written over.
Switching Ethernet Controllers (SECs) are network switches that implement the Ethernet switching protocol. According to the protocol, the Ethernet network (cabling and Ethernet ports) operates at 10 Megabits per second. Switches which operate at the desired speed of 10 Megabits per second are known as providing “full-wire” throughput.
In U.S. patent application Ser. No. 08/790,155, filed Jan. 28, 1997, and incorporated herein by reference, in order to optimize through-put time, communication between SECs attempts to utilize the bus as little as possible so that the bus will be available as soon as a SEC wants to utilize it. Therefore, each SEC includes a write-only bus communication unit which transfers the packets out of the SEC by only writing to the bus. Thus, packets enter each SEC by having been written therein from other SECs and not by reading them in, since read operations utilize the bus for significant amounts of time compared to write operations. Having the bus available generally whenever a SEC needs it helps to provide the full-wire throughput.
However, when many SECs write to the same bus, the throughput is limited by the speed of the bus.
A crossbar for communicating with at least one device, the crossbar comprises N ports. Each one of the N ports comprises a link logic unit to receive messages and data from a respective device, N−1 output buffers each corresponding to another one of the N−1 ports and a port arbiter to select one of the N−1 output buffers to output data to the respective device. The stored data is transferred to the corresponding output buffer of a selected one of the other one of the N ports.
In other features, the link logic unit determines a type of message from the respective device. If the type of message is a local link message, a port function is performed and the message is not transferred. If the type of message is a switch link message, the message and the data are transferred. The device comprises one of a switch and a second crossbar. The N−1 output buffers comprises a device table register to store a device number if the device comprises the switch. The N−1 output buffers comprises a device table register to store device numbers of devices connected to the second crossbar.
In still other features, if the device is unable to receive data, the device provides a message to the link logic unit. The link logic unit signals the arbiter to inhibit communication to the device. If an nth one of the of the N−1 output buffers is at least a predetermined capacity, a signal is sent to the nth one of the port arbiter of the other one of the N−1 ports to inhibit further transmission.
The present invention will be understood and appreciated more fully from the following detailed description taken in conjunction with the drawings in which:
Reference is now made to
As further illustrated in
As shown in
Each port 16, and the elements which it comprises, are dedicated to an associated switch 12 or cross bar 10, and are responsible for all communication with its associated switch 12 or crossbar 10. As an example, the elements which comprise port 16A are associated with switch 12A and are responsible for switch 12A's communication: link logic unit 18A receives and directs messages and data from switch 12A, and performs various port functions which will be described in more detail hereinbelow; input FIFO buffer 20A receives and stores data packets sent from switch 12A; output buffers 22 at port 16A receive and store data packets sent to switch 12A; and, port transmit arbiter 26A sends messages and data to switch 12A.
Port to port communication is made via point-to-point connection between input buffers 20 and their associated output buffers 22. As an example, and as represented in
Input buffer 20A transfers data packets via output buffers 22A to ports 16B, 16C and 16D. Additionally, though not represented in the Figures, input buffer 20B is connected to output buffers 22B, located at ports 16A, 16C and 16D, and so on. This point-to-point connection allows simultaneous non-collision data transfers from the input buffers 20 to their dedicated output buffers 22, and hence simultaneous communication between ports 16.
Communication starts at switch 12, which writes a link message and/or an accompanying data packet into its associated link logic unit 18. Link messages are either sent alone or interleaved with data packets. Two major types of link messages are possible; switch link messages to be transferred to another switch, and local link messages indicating the status of the associated switch 12. Switch link messages transferred with an accompanying data packet contain information concerning the associated data packet.
Logic unit 18 reads the first bit of the link message, which identifies the type of link message being transmitted. As an example, if the first bit is set or 1, as an example, then logic unit 18 recognizes the message as a local link message, does not transfer the message, and proceeds to perform port functions which will be described in more detail hereinbelow. If the first bit is not set or is 0, then the logic unit 18 recognizes the link message as being a switch link message and transfers the message and the associated packet to the input buffer 20.
As an example, switch 12A sends a link message and data packet to logic unit 18A; logic unit 18A identifies the link message as a switch link message and transfers both the switch link message and the data packet to input buffer 20A.
Via the direct point-to-point connections noted hereinabove, the input buffer 20 on port 16 transfers the link message and data packet to its associated output buffers 22 located on the other ports 16 on cross bar 10. As an example, input buffer 20A transfers the message and packet to output buffers 22A located on ports 16B, 16C and 16D.
Each port 16 is identified by a port address, and each device, connected to one of the switches, is identified by its own device address or number. Each output buffer 22 comprises a device table register 24, coupled thereto. When port 16 is linked to a switch 12, register 24 logically holds the device number of a device linked to that specific port 16. When port 16 is linked to a cross bar 10, register 24 logically holds the device numbers of all the devices connected to the cross bar 10 linked to that specific port 16
The device number is also logically contained within the switch link message. When the switch link message and data packets are received at the port 16, the packet is first received by the device table register 24. If the device number data in the switch link message is included in the receiving register 24, the message and packet is written into the coupled output buffer 22; if the device number data in the switch link message is not included in the receiving register 24, then the register 24 simply ignores the message, and does not receive it.
As an example, the switch link message and its associated data packet of the previous example are intended for transfer to a device linked to port 16B. Output buffers 22A located on ports 16B, 16C and 16D receive the message and the packet. The registers 24A at ports 16C and 16D do not include the device number contained in the switch message, and thus ignore the message. The register 24A coupled to output buffer 22A at port 16B does include the device number, and hence writes the message and packet into output buffer 22A at port 16B. Output buffer 22A at port 16B then transfers the message and packet to the port transmit arbiter 26B, which transfers the message and packet via link 14B to switch 12B, and onto its eventual destination.
It is common that some switches 12 and their dedicated ports 16 are busier than others. For example, if switch 12A is excessively busy, the memory in switch 12A may become too full to receive more data packets. In such a case, the present invention implements flow control which ensures that no data is lost during the temporary backup.
In this situation, if switch 12A is full, it notifies such to logic unit 18A. Logic unit 18A then signals to port transmit arbiter 26A to cut off the data flow via link 14A to switch 12A. As a result, switch 12A does not receive any more data packets, however, it does continue to send link messages and/or data packets to input buffer 20A. Once the memory in switch 12A is cleared, switch 12A sends a message to logic unit 18A to re-open up the link 14A and re-allow data packet transfer to the switch.
While the link 14A to switch 12A is closed, all data packets sent to switch 12A are temporarily stored in the output buffers 22B, 22C, and 22D, located at port 16A. If port 16A continues to be busy, the temporarily stored data messages may backup in one of the output buffers 22 at port 16A, as an example, output buffer 22D at port 16A (
Each output buffer 22 comprises an almost full threshold 23, which when crossed, signifies that the output buffer 22 has become almost full and can not receive anymore data packets. Similarly, each output buffer 22 comprises an almost empty threshold 21, which when crossed, signifies that output buffer 22 has become almost empty and can again receive data packets.
In accordance with a preferred embodiment of the present invention, and as per the example from above, the data in output buffer 22D crosses almost full threshold 23. As indicated in
Referring now to
As per the above example, port transmit arbiter 26D notifies switch arbiter 28D of switch 12D (arrow 27) to stop transferring data to switch 12A. Arbiter 28D stops transfer of data to switch 12A, however, still transfers data to other switches, as indicated by arrows 25.
When the temporarily stored data packets at output buffer 22D (at port 16A) have cleared-out, and have crossed the almost empty threshold 21, the output buffer 22D notifies sends a message to bar arbiter 26D (at port 16D) notifying it that switch 12 A is almost empty. Port transmit arbiter 26D resumes sending data packets to port 16A.
For the purpose of temporarily storing data packets while cross bar arbiter 26 has halted traffic, and as shown in
When port transmit arbiter 26D notifies arbiter 28D of switch 12D to stop sending data packets to switch 12A, switch arbiter 28D so indicates to the DMA unit 30A, DMA unit 30A stops transferring data from FIFO buffer 32A to switch 12A. However, the remaining DMA units 30 remain active, sending data through port 16D to the other switches 12.
While the outgoing data traffic from switch 12 is closed, data packets intended for transfer are temporarily stored in switch buffers 32. As noted above, each DMA 30 is linked to an associated switch buffer 32. Additionally, each DMA 30 is dedicated to one or more different ports 16. Generally each DMA 30 is dedicated to 2 ports, as example, ports 16A and 16D, however, DMA 30 could be dedicated to any number of ports depending on the switch's load.
When arbiter 28 receives a message to close off outflow of data from switch 12 to an almost full port 16, only the outflow from the DMA 30 associated with the “full” port 16 is cut off. The DMA's 30 associated with the other ports 16 remain active. Consequently, the only affected data flows are those of the associated ports and not the entire outgoing data flow from switch 12.
Hence, in contrast to prior art network systems which required a total data flow halt upon collision from backups, the flow control system of the present invention allows data flow to continue, and stems only that traffic affected by the backup.
Reference is now made to
Switch 12A transfers (step 110) a link message and a data packet to logic unit 18A. The logic unit 18A identifies (step 112) the link message as a switch link message and writes (step 114) the link message and data packet into input buffer 20A.
The input buffer 20A transfers (step 116) the link message and data packet to the registers 24A at ports 16B, 16C and 16D. The registers 24A at ports 16B, 16C and 16D receive (step 118) the link message and the data packet transferred from switch 12A. Registers 24A at ports 16C and 16D do not recognize (step 120) the device number in the link message and ignore the message and the data packet. The register 24A at port 16B, recognizes (step 122) the device number in the link message, and the link message and data packet are written (step 124) into the output buffer 22A at port 16B.
If switch 12B is not full, output buffer 22A at port 16B transfers (step 126) the data to port transmit arbiter 26B, which transfers (step 128) the message and data via link 14B to switch 12B.
If switch 12B becomes full, it transfers (step 130,
When output buffer 22A at port 16B reaches (step 136) almost fullness threshold 23, it notifies (step 138) such to port transmit arbiter 26A at port 16A. Arbiter 26A notifies (step 140) switch arbiter 28A at switch 12A not to send anymore data to port 16B until further notice. Switch 12A indicates (step 142) to its DMA 30B (on switch 12A) to stop sending to switch 12B, and switch buffer 32B (on switch 12A and dedicated to port 16B) temporarily stores (step 143) unsendable messages and data.
When switch 12B is capable of receiving again, it sends (step 144) a local link message to logic unit 18B indicating that it is now open to receive. Logic unit 18B indicates (step 146) to port transmit arbiter 26B to reopen inflow to switch 12B. Output buffer 22A at port 16B restarts transmission (step 148). When the output buffer 22A at port 16B reaches an empty threshold 21 (step 150) it notifies such to port transmit arbiter 26A at port 16A.
Port transmit arbiter 26A notifies (step 152) switch arbiter 28A at switch 12A to resume transfer of data to port 16B. Switch A indicates (step 154) to DMA 30B to resume sending data. Switch buffer 32B (on switch 12A and dedicated to port 16B) resumes (step 156) transfer of messages and data to switch 12B.
Reference is now made to
Individual bus 14 is a 17 bit, point-to-point bus and comprises a clock signal 210, a command bit signal 212 and 16 bits of data signal 214. The 16 bit data 214 transfers either a 16 bit link message or data packets.
As noted hereinabove, each link 14 provides the connection and communication between one cross bar 10 and one switch 12, and transfers link messages and data packets therebetween. As additionally noted hereinabove, link messages 214 are transferred either alone or accompanied by data packets, and switch link messages transferred with data packets are used to communicate information about the associated data packet.
Clock 210 functions in a manner similar to network clocks known in the art, and as such will not be described in detail herein. Command bit 212 is a one cycle command word and data bits 214, depending on the command message, comprises between 0 to 33 cycles of 16 bit data words. In accordance with a preferred embodiment of the present invention, command messages signal the commencement or end of a data packet. When not transmitting a command message, command bit 212 transmits an idle signal.
Referring now to
Referring now to
As noted hereinabove, the link message bit 220 is either set or not set, and is used to signify either a local link message or a switch link message, respectively. The high address 222 is provided for PCI address mapping and used for communication between switches 12 and PCIs connected to the network. The device number 224 identifies the number of the device designated to receive the link message and/or data packet. The message type 226, as described in U.S. patent application Ser. No. 08/790,155, which is incorporated herein by reference, relays messaging protocol between switches 12.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US4922348||Feb 10, 1989||May 1, 1990||American Telephone And Telegraph Company||Facsimile service|
|US5179669||Aug 22, 1988||Jan 12, 1993||At&T Bell Laboratories||Multiprocessor interconnection and access arbitration arrangement|
|US5923660||Jan 28, 1997||Jul 13, 1999||Galileo Technologies Ltd.||Switching ethernet controller|
|US6032205||Mar 6, 1998||Feb 29, 2000||Hitachi, Ltd.||Crossbar switch system for always transferring normal messages and selectively transferring broadcast messages from input buffer to output buffer when it has sufficient space respectively|
|US6067298 *||Oct 22, 1997||May 23, 2000||Nec Corporation||ATM switching system which separates services classes and uses a code switching section and back pressure signals|
|US6157967||Dec 30, 1997||Dec 5, 2000||Tandem Computer Incorporated||Method of data communication flow control in a data processing system using busy/ready commands|
|US6192422||Mar 6, 1998||Feb 20, 2001||Alcatel Internetworking, Inc.||Repeater with flow control device transmitting congestion indication data from output port buffer to associated network node upon port input buffer crossing threshold level|
|US6205542||Jan 14, 1999||Mar 20, 2001||Intel Corporation||Processor pipeline including replay|
|US6226292||Mar 19, 1998||May 1, 2001||3Com Corporation||Frame replication in a network switch for multi-port frame forwarding|
|US6314501||Dec 18, 1998||Nov 6, 2001||Unisys Corporation||Computer system and method for operating multiple operating systems in different partitions of the computer system and for allowing the different partitions to communicate with one another through shared memory|
|US6353618||Feb 4, 1999||Mar 5, 2002||Alcatel Canada Inc.||Method and apparatus for controlling traffic flows in a packet-switched network|
|US6370148||Jul 6, 1998||Apr 9, 2002||International Business Machines Corporation||Data communications|
|US6388993||Jun 11, 1998||May 14, 2002||Samsung Electronics Co., Ltd.||ATM switch and a method for determining buffer threshold|
|US6614796||Nov 19, 1998||Sep 2, 2003||Gadzoox Networks, Inc,||Fibre channel arbitrated loop bufferless switch circuitry to increase bandwidth without significant increase in cost|
|US6967962||Jul 7, 1999||Nov 22, 2005||Marvell Semiconductor Israel Ltd.||Linking crossbar controller|
|US7477652||Jun 15, 2005||Jan 13, 2009||Marvell Israel (M.I.S.L.) Ltd.||Linking cross bar controller|
|1||Newton, Newton's Telecom Dictionary, Oct. 1998, Telecom Books, p. 234.|
|2||Newton, Newton's Telecom Dictionary, U.S. Telecom Books, 1998, p. 197.|
|U.S. Classification||370/413, 370/416, 370/428|
|International Classification||H04L12/28, H04L12/46, H04L, G06F13/38, H04L12/935, H04L12/933, H04L12/931, H04L12/947|
|Cooperative Classification||H04L49/15, H04L49/505, H04L49/3018, H04L49/501, H04L12/46, H04L49/25, H04L49/3009, H04L49/3027, H04L49/101, G06F13/385|
|European Classification||G06F13/38A2, H04L49/10A, H04L49/15|